Buccaneers at Cowboys: Dallas Is the Pick. The Spread Is Close.

All five winner models pick Dallas tonight. Together, they give the Cowboys 59.8%, well below the 79.2% implied by Wednesday evening’s normalized market prices. The separate spread forecast favors Dallas -8.5 at 55.5%, while the point model projects a margin just below that line.

Tampa Bay (0–4) visits Dallas (2–2) at AT&T Stadium in Arlington on Thursday, October 8, at 7:15 p.m. Central. Prime Video carries the game nationally, with FOX 4 showing it in Dallas–Fort Worth. The Cowboys’ viewing guide confirms the details.

Thursday’s operations verified the saved forecasts and their arithmetic. They did not produce a new forecast or a Thursday odds capture. The personnel reporting below is current to this afternoon’s review; every quoted price is explicitly Wednesday’s saved observation.

A rookie’s first road start meets two depleted defenses

Baker Mayfield is out with a right-thumb injury, leaving Jalon Daniels expected to make his second NFL start and first on the road. In his 17–14 loss to Green Bay, Daniels threw for 148 yards, one touchdown and two interceptions, adding 55 rushing yards. Dallas has produced four sacks and no interceptions through four games, according to Tampa Bay’s game preview. Daniels’ mobility and mistakes both matter against that defense.

Tampa Bay’s absences extend beyond quarterback. Antoine Winfield Jr., Benjamin Morrison and SirVocea Dennis are also out, according to the Bucs’ final injury report.

They will be missing against an offense coming off a 34–30 win at Houston. Dak Prescott threw for 335 yards, and CeeDee Lamb caught 17 passes for 189 yards, per the Cowboys’ recap. That production gives the Bucs’ secondary absences particular relevance tonight.

Dallas has its own absences: Cobie Durant, DeMarvion Overshown and Drew Shelton are out, while Jonathan Mingo was listed questionable in the NFL’s injury report. Those are reported designations, not a confirmed game-day inactive list.

There is a newer update on Tyler Smith. He was questionable Wednesday, but Dallas has since activated him from injured reserve and says he will suit up against Tampa Bay. The Cowboys’ Thursday announcement is the relevant update for his planned return; activation alone would not have established that.

Those reports frame the matchup, including Smith’s planned return to the offensive line. They have not changed the frozen forecasts, and I have not calculated a new injury adjustment from them.

Five Dallas votes do not make an overwhelming favorite

The full Week 5 preview explains the single issued forecast. It was generated at 3:17 a.m. Central Thursday and separately verified and frozen at 3:39:58 a.m., before tonight’s kickoff.

Every number in this table is Dallas’s probability of winning outright.

Logistic supplies the strongest Dallas opinion. XGBoost barely separates the teams. All five votes point the same way, but their probabilities range across 18.5 percentage points. The ensemble combines those strengths with fitted weights close to one-fifth each, rather than treating unanimity as an extra confidence boost.

The result leaves 40.2% for a Tampa Bay win. The forecast uses T = 1, preserving the raw blend after the proposed calibration failed its holdout checks. There is no second sharpened version for this game, and the unchanged raw output remains uncalibrated.

The saved market was much more certain about Dallas

The latest retained BetMGM capture is Wednesday, October 7, at 8:59 p.m. Central: Dallas -475, Tampa Bay +360, Dallas -8.5 and a total of 48.

Normalizing that moneyline pair to remove its quoted margin gives Dallas 79.2%, compared with the model’s 59.8%. The gap is 19.4 percentage points, calculated before rounding. Both pick Dallas; they disagree substantially about the chance of an upset.

At the same bookmaker’s 8:34 a.m. Wednesday observation, Dallas was -450 and Tampa Bay +350. The handicap stayed -8.5 through the evening capture, while the total rose 47.5 to 48.

Those are two observed prices from the same book. They are not guaranteed opening or closing lines, and I have no verified Thursday capture to describe as a fresh market move. The gap also does not establish a betting advantage: it depends on probabilities whose calibration remains unresolved.

The point projection sits almost on the spread

The XGBoost margin and total models supply these point estimates. Team scores are derived from (total ± margin) / 2; they are separate from both the winner blend and the spread-cover probabilities.

The retained margin is 8.4567, just 0.0433 points below 8.5. Rounding it to “Dallas by 8.5” would hide which side of the benchmark it falls on. This point projection does not clear Dallas -8.5, although the difference is far too small to describe as meaningful separation.

Actual NFL scores are whole numbers: Dallas needs to win by at least nine to cover -8.5. Tampa Bay +8.5 can cover even if the Buccaneers lose by one through eight. Dallas remains the favorite in either description; picking a team against the spread is a different question from picking the winner.

The projected 48.48 total is only 0.48 above the saved line. That is a point estimate, not a probability that the game goes over.

The spread models agree; the totals models do not

Here are the individual trained market estimates and the saved aggregates. The two rows answer different questions at their exact thresholds.

The first row uses Logistic ATS; the second uses the linear totals model. All three trained spread models prefer Dallas covering, with Random Forest strongest. That separate classification blend need not match the standalone margin projection.

Totals contain a larger disagreement. The linear model projects 56.27 points, versus XGBoost’s 48.48, and gives the over a much stronger probability. Random Forest slightly prefers the under. The totals aggregate also retains neutral 50% contributions that pull the combined estimate toward even; those are not additional trained totals models.

At exactly 48 points, the saved total pushes. The export provides no separate push probability, so its over/under numbers should not be presented as a complete three-outcome distribution. If the offered spread or total changes, these saved probabilities cannot simply be attached to the new line.

What tonight will put on the record

My saved selections are Dallas to win, Dallas -8.5 to cover, and Over 48. The first is a 59.8% call; the latter two are more modest 55.5% and 54.9% estimates. Unanimous winner votes do not make all three outcomes likely to land together.

The follow-up will grade the winner blend and every component, compare the projected points and margin with the actual score, and settle the spread and total at these saved thresholds. Pushes stay separate. As the Week 4 review showed, correct winners, probability scores and spread results can tell different stories.

Tonight opens a fifteen-game Week 5 slate. The forecast and its limitations are recorded before the result; the remaining fourteen games stay on the weekly board.

Source note: Game 2026_05_TB_DAL uses the frozen Week 5 forecast, Wednesday’s saved BetMGM captures and the Thursday writer handoff committed at prediction-repository revision 185a754. Official football reporting was checked Thursday afternoon; final game-day inactives were not verified. Companion editorial evidence retains full-precision probabilities, score calculations, timestamps and source hashes. Preparing this post did not refit models, acquire new odds or change grading records.